Pileus Git - ~andy/linux/blob - drivers/md/bcache/btree.c

   1 /*
   2  * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
   3  *
   4  * Uses a block device as cache for other block devices; optimized for SSDs.
   5  * All allocation is done in buckets, which should match the erase block size
   6  * of the device.
   7  *
   8  * Buckets containing cached data are kept on a heap sorted by priority;
   9  * bucket priority is increased on cache hit, and periodically all the buckets
  10  * on the heap have their priority scaled down. This currently is just used as
  11  * an LRU but in the future should allow for more intelligent heuristics.
  12  *
  13  * Buckets have an 8 bit counter; freeing is accomplished by incrementing the
  14  * counter. Garbage collection is used to remove stale pointers.
  15  *
  16  * Indexing is done via a btree; nodes are not necessarily fully sorted, rather
  17  * as keys are inserted we only sort the pages that have not yet been written.
  18  * When garbage collection is run, we resort the entire node.
  19  *
  20  * All configuration is done via sysfs; see Documentation/bcache.txt.
  21  */
  22
  23 #include "bcache.h"
  24 #include "btree.h"
  25 #include "debug.h"
  26 #include "request.h"
  27 #include "writeback.h"
  28
  29 #include <linux/slab.h>
  30 #include <linux/bitops.h>
  31 #include <linux/hash.h>
  32 #include <linux/prefetch.h>
  33 #include <linux/random.h>
  34 #include <linux/rcupdate.h>
  35 #include <trace/events/bcache.h>
  36
  37 /*
  38  * Todo:
  39  * register_bcache: Return errors out to userspace correctly
  40  *
  41  * Writeback: don't undirty key until after a cache flush
  42  *
  43  * Create an iterator for key pointers
  44  *
  45  * On btree write error, mark bucket such that it won't be freed from the cache
  46  *
  47  * Journalling:
  48  *   Check for bad keys in replay
  49  *   Propagate barriers
  50  *   Refcount journal entries in journal_replay
  51  *
  52  * Garbage collection:
  53  *   Finish incremental gc
  54  *   Gc should free old UUIDs, data for invalid UUIDs
  55  *
  56  * Provide a way to list backing device UUIDs we have data cached for, and
  57  * probably how long it's been since we've seen them, and a way to invalidate
  58  * dirty data for devices that will never be attached again
  59  *
  60  * Keep 1 min/5 min/15 min statistics of how busy a block device has been, so
  61  * that based on that and how much dirty data we have we can keep writeback
  62  * from being starved
  63  *
  64  * Add a tracepoint or somesuch to watch for writeback starvation
  65  *
  66  * When btree depth > 1 and splitting an interior node, we have to make sure
  67  * alloc_bucket() cannot fail. This should be true but is not completely
  68  * obvious.
  69  *
  70  * Make sure all allocations get charged to the root cgroup
  71  *
  72  * Plugging?
  73  *
  74  * If data write is less than hard sector size of ssd, round up offset in open
  75  * bucket to the next whole sector
  76  *
  77  * Also lookup by cgroup in get_open_bucket()
  78  *
  79  * Superblock needs to be fleshed out for multiple cache devices
  80  *
  81  * Add a sysfs tunable for the number of writeback IOs in flight
  82  *
  83  * Add a sysfs tunable for the number of open data buckets
  84  *
  85  * IO tracking: Can we track when one process is doing io on behalf of another?
  86  * IO tracking: Don't use just an average, weigh more recent stuff higher
  87  *
  88  * Test module load/unload
  89  */
  90
  91 static const char * const op_types[] = {
  92         "insert", "replace"
  93 };
  94
  95 static const char *op_type(struct btree_op *op)
  96 {
  97         return op_types[op->type];
  98 }
  99
 100 #define MAX_NEED_GC             64
 101 #define MAX_SAVE_PRIO           72
 102
 103 #define PTR_DIRTY_BIT           (((uint64_t) 1 << 36))
 104
 105 #define PTR_HASH(c, k)                                                  \
 106         (((k)->ptr[0] >> c->bucket_bits) | PTR_GEN(k, 0))
 107
 108 struct workqueue_struct *bch_gc_wq;
 109 static struct workqueue_struct *btree_io_wq;
 110
 111 void bch_btree_op_init_stack(struct btree_op *op)
 112 {
 113         memset(op, 0, sizeof(struct btree_op));
 114         closure_init_stack(&op->cl);
 115         op->lock = -1;
 116         bch_keylist_init(&op->keys);
 117 }
 118
 119 /* Btree key manipulation */
 120
 121 static void bkey_put(struct cache_set *c, struct bkey *k, int level)
 122 {
 123         if ((level && KEY_OFFSET(k)) || !level)
 124                 __bkey_put(c, k);
 125 }
 126
 127 /* Btree IO */
 128
 129 static uint64_t btree_csum_set(struct btree *b, struct bset *i)
 130 {
 131         uint64_t crc = b->key.ptr[0];
 132         void *data = (void *) i + 8, *end = end(i);
 133
 134         crc = bch_crc64_update(crc, data, end - data);
 135         return crc ^ 0xffffffffffffffffULL;
 136 }
 137
 138 static void bch_btree_node_read_done(struct btree *b)
 139 {
 140         const char *err = "bad btree header";
 141         struct bset *i = b->sets[0].data;
 142         struct btree_iter *iter;
 143
 144         iter = mempool_alloc(b->c->fill_iter, GFP_NOWAIT);
 145         iter->size = b->c->sb.bucket_size / b->c->sb.block_size;
 146         iter->used = 0;
 147
 148         if (!i->seq)
 149                 goto err;
 150
 151         for (;
 152              b->written < btree_blocks(b) && i->seq == b->sets[0].data->seq;
 153              i = write_block(b)) {
 154                 err = "unsupported bset version";
 155                 if (i->version > BCACHE_BSET_VERSION)
 156                         goto err;
 157
 158                 err = "bad btree header";
 159                 if (b->written + set_blocks(i, b->c) > btree_blocks(b))
 160                         goto err;
 161
 162                 err = "bad magic";
 163                 if (i->magic != bset_magic(b->c))
 164                         goto err;
 165
 166                 err = "bad checksum";
 167                 switch (i->version) {
 168                 case 0:
 169                         if (i->csum != csum_set(i))
 170                                 goto err;
 171                         break;
 172                 case BCACHE_BSET_VERSION:
 173                         if (i->csum != btree_csum_set(b, i))
 174                                 goto err;
 175                         break;
 176                 }
 177
 178                 err = "empty set";
 179                 if (i != b->sets[0].data && !i->keys)
 180                         goto err;
 181
 182                 bch_btree_iter_push(iter, i->start, end(i));
 183
 184                 b->written += set_blocks(i, b->c);
 185         }
 186
 187         err = "corrupted btree";
 188         for (i = write_block(b);
 189              index(i, b) < btree_blocks(b);
 190              i = ((void *) i) + block_bytes(b->c))
 191                 if (i->seq == b->sets[0].data->seq)
 192                         goto err;
 193
 194         bch_btree_sort_and_fix_extents(b, iter);
 195
 196         i = b->sets[0].data;
 197         err = "short btree key";
 198         if (b->sets[0].size &&
 199             bkey_cmp(&b->key, &b->sets[0].end) < 0)
 200                 goto err;
 201
 202         if (b->written < btree_blocks(b))
 203                 bch_bset_init_next(b);
 204 out:
 205         mempool_free(iter, b->c->fill_iter);
 206         return;
 207 err:
 208         set_btree_node_io_error(b);
 209         bch_cache_set_error(b->c, "%s at bucket %zu, block %zu, %u keys",
 210                             err, PTR_BUCKET_NR(b->c, &b->key, 0),
 211                             index(i, b), i->keys);
 212         goto out;
 213 }
 214
 215 static void btree_node_read_endio(struct bio *bio, int error)
 216 {
 217         struct closure *cl = bio->bi_private;
 218         closure_put(cl);
 219 }
 220
 221 void bch_btree_node_read(struct btree *b)
 222 {
 223         uint64_t start_time = local_clock();
 224         struct closure cl;
 225         struct bio *bio;
 226
 227         trace_bcache_btree_read(b);
 228
 229         closure_init_stack(&cl);
 230
 231         bio = bch_bbio_alloc(b->c);
 232         bio->bi_rw      = REQ_META|READ_SYNC;
 233         bio->bi_size    = KEY_SIZE(&b->key) << 9;
 234         bio->bi_end_io  = btree_node_read_endio;
 235         bio->bi_private = &cl;
 236
 237         bch_bio_map(bio, b->sets[0].data);
 238
 239         bch_submit_bbio(bio, b->c, &b->key, 0);
 240         closure_sync(&cl);
 241
 242         if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
 243                 set_btree_node_io_error(b);
 244
 245         bch_bbio_free(bio, b->c);
 246
 247         if (btree_node_io_error(b))
 248                 goto err;
 249
 250         bch_btree_node_read_done(b);
 251
 252         spin_lock(&b->c->btree_read_time_lock);
 253         bch_time_stats_update(&b->c->btree_read_time, start_time);
 254         spin_unlock(&b->c->btree_read_time_lock);
 255
 256         return;
 257 err:
 258         bch_cache_set_error(b->c, "io error reading bucket %zu",
 259                             PTR_BUCKET_NR(b->c, &b->key, 0));
 260 }
 261
 262 static void btree_complete_write(struct btree *b, struct btree_write *w)
 263 {
 264         if (w->prio_blocked &&
 265             !atomic_sub_return(w->prio_blocked, &b->c->prio_blocked))
 266                 wake_up_allocators(b->c);
 267
 268         if (w->journal) {
 269                 atomic_dec_bug(w->journal);
 270                 __closure_wake_up(&b->c->journal.wait);
 271         }
 272
 273         w->prio_blocked = 0;
 274         w->journal      = NULL;
 275 }
 276
 277 static void __btree_node_write_done(struct closure *cl)
 278 {
 279         struct btree *b = container_of(cl, struct btree, io.cl);
 280         struct btree_write *w = btree_prev_write(b);
 281
 282         bch_bbio_free(b->bio, b->c);
 283         b->bio = NULL;
 284         btree_complete_write(b, w);
 285
 286         if (btree_node_dirty(b))
 287                 queue_delayed_work(btree_io_wq, &b->work,
 288                                    msecs_to_jiffies(30000));
 289
 290         closure_return(cl);
 291 }
 292
 293 static void btree_node_write_done(struct closure *cl)
 294 {
 295         struct btree *b = container_of(cl, struct btree, io.cl);
 296         struct bio_vec *bv;
 297         int n;
 298
 299         __bio_for_each_segment(bv, b->bio, n, 0)
 300                 __free_page(bv->bv_page);
 301
 302         __btree_node_write_done(cl);
 303 }
 304
 305 static void btree_node_write_endio(struct bio *bio, int error)
 306 {
 307         struct closure *cl = bio->bi_private;
 308         struct btree *b = container_of(cl, struct btree, io.cl);
 309
 310         if (error)
 311                 set_btree_node_io_error(b);
 312
 313         bch_bbio_count_io_errors(b->c, bio, error, "writing btree");
 314         closure_put(cl);
 315 }
 316
 317 static void do_btree_node_write(struct btree *b)
 318 {
 319         struct closure *cl = &b->io.cl;
 320         struct bset *i = b->sets[b->nsets].data;
 321         BKEY_PADDED(key) k;
 322
 323         i->version      = BCACHE_BSET_VERSION;
 324         i->csum         = btree_csum_set(b, i);
 325
 326         BUG_ON(b->bio);
 327         b->bio = bch_bbio_alloc(b->c);
 328
 329         b->bio->bi_end_io       = btree_node_write_endio;
 330         b->bio->bi_private      = &b->io.cl;
 331         b->bio->bi_rw           = REQ_META|WRITE_SYNC|REQ_FUA;
 332         b->bio->bi_size         = set_blocks(i, b->c) * block_bytes(b->c);
 333         bch_bio_map(b->bio, i);
 334
 335         /*
 336          * If we're appending to a leaf node, we don't technically need FUA -
 337          * this write just needs to be persisted before the next journal write,
 338          * which will be marked FLUSH|FUA.
 339          *
 340          * Similarly if we're writing a new btree root - the pointer is going to
 341          * be in the next journal entry.
 342          *
 343          * But if we're writing a new btree node (that isn't a root) or
 344          * appending to a non leaf btree node, we need either FUA or a flush
 345          * when we write the parent with the new pointer. FUA is cheaper than a
 346          * flush, and writes appending to leaf nodes aren't blocking anything so
 347          * just make all btree node writes FUA to keep things sane.
 348          */
 349
 350         bkey_copy(&k.key, &b->key);
 351         SET_PTR_OFFSET(&k.key, 0, PTR_OFFSET(&k.key, 0) + bset_offset(b, i));
 352
 353         if (!bio_alloc_pages(b->bio, GFP_NOIO)) {
 354                 int j;
 355                 struct bio_vec *bv;
 356                 void *base = (void *) ((unsigned long) i & ~(PAGE_SIZE - 1));
 357
 358                 bio_for_each_segment(bv, b->bio, j)
 359                         memcpy(page_address(bv->bv_page),
 360                                base + j * PAGE_SIZE, PAGE_SIZE);
 361
 362                 bch_submit_bbio(b->bio, b->c, &k.key, 0);
 363
 364                 continue_at(cl, btree_node_write_done, NULL);
 365         } else {
 366                 b->bio->bi_vcnt = 0;
 367                 bch_bio_map(b->bio, i);
 368
 369                 bch_submit_bbio(b->bio, b->c, &k.key, 0);
 370
 371                 closure_sync(cl);
 372                 __btree_node_write_done(cl);
 373         }
 374 }
 375
 376 void bch_btree_node_write(struct btree *b, struct closure *parent)
 377 {
 378         struct bset *i = b->sets[b->nsets].data;
 379
 380         trace_bcache_btree_write(b);
 381
 382         BUG_ON(current->bio_list);
 383         BUG_ON(b->written >= btree_blocks(b));
 384         BUG_ON(b->written && !i->keys);
 385         BUG_ON(b->sets->data->seq != i->seq);
 386         bch_check_key_order(b, i);
 387
 388         cancel_delayed_work(&b->work);
 389
 390         /* If caller isn't waiting for write, parent refcount is cache set */
 391         closure_lock(&b->io, parent ?: &b->c->cl);
 392
 393         clear_bit(BTREE_NODE_dirty,      &b->flags);
 394         change_bit(BTREE_NODE_write_idx, &b->flags);
 395
 396         do_btree_node_write(b);
 397
 398         b->written += set_blocks(i, b->c);
 399         atomic_long_add(set_blocks(i, b->c) * b->c->sb.block_size,
 400                         &PTR_CACHE(b->c, &b->key, 0)->btree_sectors_written);
 401
 402         bch_btree_sort_lazy(b);
 403
 404         if (b->written < btree_blocks(b))
 405                 bch_bset_init_next(b);
 406 }
 407
 408 static void btree_node_write_work(struct work_struct *w)
 409 {
 410         struct btree *b = container_of(to_delayed_work(w), struct btree, work);
 411
 412         rw_lock(true, b, b->level);
 413
 414         if (btree_node_dirty(b))
 415                 bch_btree_node_write(b, NULL);
 416         rw_unlock(true, b);
 417 }
 418
 419 static void bch_btree_leaf_dirty(struct btree *b, struct btree_op *op)
 420 {
 421         struct bset *i = b->sets[b->nsets].data;
 422         struct btree_write *w = btree_current_write(b);
 423
 424         BUG_ON(!b->written);
 425         BUG_ON(!i->keys);
 426
 427         if (!btree_node_dirty(b))
 428                 queue_delayed_work(btree_io_wq, &b->work, 30 * HZ);
 429
 430         set_btree_node_dirty(b);
 431
 432         if (op && op->journal) {
 433                 if (w->journal &&
 434                     journal_pin_cmp(b->c, w, op)) {
 435                         atomic_dec_bug(w->journal);
 436                         w->journal = NULL;
 437                 }
 438
 439                 if (!w->journal) {
 440                         w->journal = op->journal;
 441                         atomic_inc(w->journal);
 442                 }
 443         }
 444
 445         /* Force write if set is too big */
 446         if (set_bytes(i) > PAGE_SIZE - 48 &&
 447             !current->bio_list)
 448                 bch_btree_node_write(b, NULL);
 449 }
 450
 451 /*
 452  * Btree in memory cache - allocation/freeing
 453  * mca -> memory cache
 454  */
 455
 456 static void mca_reinit(struct btree *b)
 457 {
 458         unsigned i;
 459
 460         b->flags        = 0;
 461         b->written      = 0;
 462         b->nsets        = 0;
 463
 464         for (i = 0; i < MAX_BSETS; i++)
 465                 b->sets[i].size = 0;
 466         /*
 467          * Second loop starts at 1 because b->sets[0]->data is the memory we
 468          * allocated
 469          */
 470         for (i = 1; i < MAX_BSETS; i++)
 471                 b->sets[i].data = NULL;
 472 }
 473
 474 #define mca_reserve(c)  (((c->root && c->root->level)           \
 475                           ? c->root->level : 1) * 8 + 16)
 476 #define mca_can_free(c)                                         \
 477         max_t(int, 0, c->bucket_cache_used - mca_reserve(c))
 478
 479 static void mca_data_free(struct btree *b)
 480 {
 481         struct bset_tree *t = b->sets;
 482         BUG_ON(!closure_is_unlocked(&b->io.cl));
 483
 484         if (bset_prev_bytes(b) < PAGE_SIZE)
 485                 kfree(t->prev);
 486         else
 487                 free_pages((unsigned long) t->prev,
 488                            get_order(bset_prev_bytes(b)));
 489
 490         if (bset_tree_bytes(b) < PAGE_SIZE)
 491                 kfree(t->tree);
 492         else
 493                 free_pages((unsigned long) t->tree,
 494                            get_order(bset_tree_bytes(b)));
 495
 496         free_pages((unsigned long) t->data, b->page_order);
 497
 498         t->prev = NULL;
 499         t->tree = NULL;
 500         t->data = NULL;
 501         list_move(&b->list, &b->c->btree_cache_freed);
 502         b->c->bucket_cache_used--;
 503 }
 504
 505 static void mca_bucket_free(struct btree *b)
 506 {
 507         BUG_ON(btree_node_dirty(b));
 508
 509         b->key.ptr[0] = 0;
 510         hlist_del_init_rcu(&b->hash);
 511         list_move(&b->list, &b->c->btree_cache_freeable);
 512 }
 513
 514 static unsigned btree_order(struct bkey *k)
 515 {
 516         return ilog2(KEY_SIZE(k) / PAGE_SECTORS ?: 1);
 517 }
 518
 519 static void mca_data_alloc(struct btree *b, struct bkey *k, gfp_t gfp)
 520 {
 521         struct bset_tree *t = b->sets;
 522         BUG_ON(t->data);
 523
 524         b->page_order = max_t(unsigned,
 525                               ilog2(b->c->btree_pages),
 526                               btree_order(k));
 527
 528         t->data = (void *) __get_free_pages(gfp, b->page_order);
 529         if (!t->data)
 530                 goto err;
 531
 532         t->tree = bset_tree_bytes(b) < PAGE_SIZE
 533                 ? kmalloc(bset_tree_bytes(b), gfp)
 534                 : (void *) __get_free_pages(gfp, get_order(bset_tree_bytes(b)));
 535         if (!t->tree)
 536                 goto err;
 537
 538         t->prev = bset_prev_bytes(b) < PAGE_SIZE
 539                 ? kmalloc(bset_prev_bytes(b), gfp)
 540                 : (void *) __get_free_pages(gfp, get_order(bset_prev_bytes(b)));
 541         if (!t->prev)
 542                 goto err;
 543
 544         list_move(&b->list, &b->c->btree_cache);
 545         b->c->bucket_cache_used++;
 546         return;
 547 err:
 548         mca_data_free(b);
 549 }
 550
 551 static struct btree *mca_bucket_alloc(struct cache_set *c,
 552                                       struct bkey *k, gfp_t gfp)
 553 {
 554         struct btree *b = kzalloc(sizeof(struct btree), gfp);
 555         if (!b)
 556                 return NULL;
 557
 558         init_rwsem(&b->lock);
 559         lockdep_set_novalidate_class(&b->lock);
 560         INIT_LIST_HEAD(&b->list);
 561         INIT_DELAYED_WORK(&b->work, btree_node_write_work);
 562         b->c = c;
 563         closure_init_unlocked(&b->io);
 564
 565         mca_data_alloc(b, k, gfp);
 566         return b;
 567 }
 568
 569 static int mca_reap(struct btree *b, struct closure *cl, unsigned min_order)
 570 {
 571         lockdep_assert_held(&b->c->bucket_lock);
 572
 573         if (!down_write_trylock(&b->lock))
 574                 return -ENOMEM;
 575
 576         if (b->page_order < min_order) {
 577                 rw_unlock(true, b);
 578                 return -ENOMEM;
 579         }
 580
 581         BUG_ON(btree_node_dirty(b) && !b->sets[0].data);
 582
 583         if (cl && btree_node_dirty(b))
 584                 bch_btree_node_write(b, NULL);
 585
 586         if (cl)
 587                 closure_wait_event_async(&b->io.wait, cl,
 588                          atomic_read(&b->io.cl.remaining) == -1);
 589
 590         if (btree_node_dirty(b) ||
 591             !closure_is_unlocked(&b->io.cl) ||
 592             work_pending(&b->work.work)) {
 593                 rw_unlock(true, b);
 594                 return -EAGAIN;
 595         }
 596
 597         return 0;
 598 }
 599
 600 static unsigned long bch_mca_scan(struct shrinker *shrink,
 601                                   struct shrink_control *sc)
 602 {
 603         struct cache_set *c = container_of(shrink, struct cache_set, shrink);
 604         struct btree *b, *t;
 605         unsigned long i, nr = sc->nr_to_scan;
 606         unsigned long freed = 0;
 607
 608         if (c->shrinker_disabled)
 609                 return SHRINK_STOP;
 610
 611         if (c->try_harder)
 612                 return SHRINK_STOP;
 613
 614         /* Return -1 if we can't do anything right now */
 615         if (sc->gfp_mask & __GFP_IO)
 616                 mutex_lock(&c->bucket_lock);
 617         else if (!mutex_trylock(&c->bucket_lock))
 618                 return -1;
 619
 620         /*
 621          * It's _really_ critical that we don't free too many btree nodes - we
 622          * have to always leave ourselves a reserve. The reserve is how we
 623          * guarantee that allocating memory for a new btree node can always
 624          * succeed, so that inserting keys into the btree can always succeed and
 625          * IO can always make forward progress:
 626          */
 627         nr /= c->btree_pages;
 628         nr = min_t(unsigned long, nr, mca_can_free(c));
 629
 630         i = 0;
 631         list_for_each_entry_safe(b, t, &c->btree_cache_freeable, list) {
 632                 if (freed >= nr)
 633                         break;
 634
 635                 if (++i > 3 &&
 636                     !mca_reap(b, NULL, 0)) {
 637                         mca_data_free(b);
 638                         rw_unlock(true, b);
 639                         freed++;
 640                 }
 641         }
 642
 643         /*
 644          * Can happen right when we first start up, before we've read in any
 645          * btree nodes
 646          */
 647         if (list_empty(&c->btree_cache))
 648                 goto out;
 649
 650         for (i = 0; (nr--) && i < c->bucket_cache_used; i++) {
 651                 b = list_first_entry(&c->btree_cache, struct btree, list);
 652                 list_rotate_left(&c->btree_cache);
 653
 654                 if (!b->accessed &&
 655                     !mca_reap(b, NULL, 0)) {
 656                         mca_bucket_free(b);
 657                         mca_data_free(b);
 658                         rw_unlock(true, b);
 659                         freed++;
 660                 } else
 661                         b->accessed = 0;
 662         }
 663 out:
 664         mutex_unlock(&c->bucket_lock);
 665         return freed;
 666 }
 667
 668 static unsigned long bch_mca_count(struct shrinker *shrink,
 669                                    struct shrink_control *sc)
 670 {
 671         struct cache_set *c = container_of(shrink, struct cache_set, shrink);
 672
 673         if (c->shrinker_disabled)
 674                 return 0;
 675
 676         if (c->try_harder)
 677                 return 0;
 678
 679         return mca_can_free(c) * c->btree_pages;
 680 }
 681
 682 void bch_btree_cache_free(struct cache_set *c)
 683 {
 684         struct btree *b;
 685         struct closure cl;
 686         closure_init_stack(&cl);
 687
 688         if (c->shrink.list.next)
 689                 unregister_shrinker(&c->shrink);
 690
 691         mutex_lock(&c->bucket_lock);
 692
 693 #ifdef CONFIG_BCACHE_DEBUG
 694         if (c->verify_data)
 695                 list_move(&c->verify_data->list, &c->btree_cache);
 696 #endif
 697
 698         list_splice(&c->btree_cache_freeable,
 699                     &c->btree_cache);
 700
 701         while (!list_empty(&c->btree_cache)) {
 702                 b = list_first_entry(&c->btree_cache, struct btree, list);
 703
 704                 if (btree_node_dirty(b))
 705                         btree_complete_write(b, btree_current_write(b));
 706                 clear_bit(BTREE_NODE_dirty, &b->flags);
 707
 708                 mca_data_free(b);
 709         }
 710
 711         while (!list_empty(&c->btree_cache_freed)) {
 712                 b = list_first_entry(&c->btree_cache_freed,
 713                                      struct btree, list);
 714                 list_del(&b->list);
 715                 cancel_delayed_work_sync(&b->work);
 716                 kfree(b);
 717         }
 718
 719         mutex_unlock(&c->bucket_lock);
 720 }
 721
 722 int bch_btree_cache_alloc(struct cache_set *c)
 723 {
 724         unsigned i;
 725
 726         /* XXX: doesn't check for errors */
 727
 728         closure_init_unlocked(&c->gc);
 729
 730         for (i = 0; i < mca_reserve(c); i++)
 731                 mca_bucket_alloc(c, &ZERO_KEY, GFP_KERNEL);
 732
 733         list_splice_init(&c->btree_cache,
 734                          &c->btree_cache_freeable);
 735
 736 #ifdef CONFIG_BCACHE_DEBUG
 737         mutex_init(&c->verify_lock);
 738
 739         c->verify_data = mca_bucket_alloc(c, &ZERO_KEY, GFP_KERNEL);
 740
 741         if (c->verify_data &&
 742             c->verify_data->sets[0].data)
 743                 list_del_init(&c->verify_data->list);
 744         else
 745                 c->verify_data = NULL;
 746 #endif
 747
 748         c->shrink.count_objects = bch_mca_count;
 749         c->shrink.scan_objects = bch_mca_scan;
 750         c->shrink.seeks = 4;
 751         c->shrink.batch = c->btree_pages * 2;
 752         register_shrinker(&c->shrink);
 753
 754         return 0;
 755 }
 756
 757 /* Btree in memory cache - hash table */
 758
 759 static struct hlist_head *mca_hash(struct cache_set *c, struct bkey *k)
 760 {
 761         return &c->bucket_hash[hash_32(PTR_HASH(c, k), BUCKET_HASH_BITS)];
 762 }
 763
 764 static struct btree *mca_find(struct cache_set *c, struct bkey *k)
 765 {
 766         struct btree *b;
 767
 768         rcu_read_lock();
 769         hlist_for_each_entry_rcu(b, mca_hash(c, k), hash)
 770                 if (PTR_HASH(c, &b->key) == PTR_HASH(c, k))
 771                         goto out;
 772         b = NULL;
 773 out:
 774         rcu_read_unlock();
 775         return b;
 776 }
 777
 778 static struct btree *mca_cannibalize(struct cache_set *c, struct bkey *k,
 779                                      int level, struct closure *cl)
 780 {
 781         int ret = -ENOMEM;
 782         struct btree *i;
 783
 784         trace_bcache_btree_cache_cannibalize(c);
 785
 786         if (!cl)
 787                 return ERR_PTR(-ENOMEM);
 788
 789         /*
 790          * Trying to free up some memory - i.e. reuse some btree nodes - may
 791          * require initiating IO to flush the dirty part of the node. If we're
 792          * running under generic_make_request(), that IO will never finish and
 793          * we would deadlock. Returning -EAGAIN causes the cache lookup code to
 794          * punt to workqueue and retry.
 795          */
 796         if (current->bio_list)
 797                 return ERR_PTR(-EAGAIN);
 798
 799         if (c->try_harder && c->try_harder != cl) {
 800                 closure_wait_event_async(&c->try_wait, cl, !c->try_harder);
 801                 return ERR_PTR(-EAGAIN);
 802         }
 803
 804         c->try_harder = cl;
 805         c->try_harder_start = local_clock();
 806 retry:
 807         list_for_each_entry_reverse(i, &c->btree_cache, list) {
 808                 int r = mca_reap(i, cl, btree_order(k));
 809                 if (!r)
 810                         return i;
 811                 if (r != -ENOMEM)
 812                         ret = r;
 813         }
 814
 815         if (ret == -EAGAIN &&
 816             closure_blocking(cl)) {
 817                 mutex_unlock(&c->bucket_lock);
 818                 closure_sync(cl);
 819                 mutex_lock(&c->bucket_lock);
 820                 goto retry;
 821         }
 822
 823         return ERR_PTR(ret);
 824 }
 825
 826 /*
 827  * We can only have one thread cannibalizing other cached btree nodes at a time,
 828  * or we'll deadlock. We use an open coded mutex to ensure that, which a
 829  * cannibalize_bucket() will take. This means every time we unlock the root of
 830  * the btree, we need to release this lock if we have it held.
 831  */
 832 void bch_cannibalize_unlock(struct cache_set *c, struct closure *cl)
 833 {
 834         if (c->try_harder == cl) {
 835                 bch_time_stats_update(&c->try_harder_time, c->try_harder_start);
 836                 c->try_harder = NULL;
 837                 __closure_wake_up(&c->try_wait);
 838         }
 839 }
 840
 841 static struct btree *mca_alloc(struct cache_set *c, struct bkey *k,
 842                                int level, struct closure *cl)
 843 {
 844         struct btree *b;
 845
 846         lockdep_assert_held(&c->bucket_lock);
 847
 848         if (mca_find(c, k))
 849                 return NULL;
 850
 851         /* btree_free() doesn't free memory; it sticks the node on the end of
 852          * the list. Check if there's any freed nodes there:
 853          */
 854         list_for_each_entry(b, &c->btree_cache_freeable, list)
 855                 if (!mca_reap(b, NULL, btree_order(k)))
 856                         goto out;
 857
 858         /* We never free struct btree itself, just the memory that holds the on
 859          * disk node. Check the freed list before allocating a new one:
 860          */
 861         list_for_each_entry(b, &c->btree_cache_freed, list)
 862                 if (!mca_reap(b, NULL, 0)) {
 863                         mca_data_alloc(b, k, __GFP_NOWARN|GFP_NOIO);
 864                         if (!b->sets[0].data)
 865                                 goto err;
 866                         else
 867                                 goto out;
 868                 }
 869
 870         b = mca_bucket_alloc(c, k, __GFP_NOWARN|GFP_NOIO);
 871         if (!b)
 872                 goto err;
 873
 874         BUG_ON(!down_write_trylock(&b->lock));
 875         if (!b->sets->data)
 876                 goto err;
 877 out:
 878         BUG_ON(!closure_is_unlocked(&b->io.cl));
 879
 880         bkey_copy(&b->key, k);
 881         list_move(&b->list, &c->btree_cache);
 882         hlist_del_init_rcu(&b->hash);
 883         hlist_add_head_rcu(&b->hash, mca_hash(c, k));
 884
 885         lock_set_subclass(&b->lock.dep_map, level + 1, _THIS_IP_);
 886         b->level        = level;
 887
 888         mca_reinit(b);
 889
 890         return b;
 891 err:
 892         if (b)
 893                 rw_unlock(true, b);
 894
 895         b = mca_cannibalize(c, k, level, cl);
 896         if (!IS_ERR(b))
 897                 goto out;
 898
 899         return b;
 900 }
 901
 902 /**
 903  * bch_btree_node_get - find a btree node in the cache and lock it, reading it
 904  * in from disk if necessary.
 905  *
 906  * If IO is necessary, it uses the closure embedded in struct btree_op to wait;
 907  * if that closure is in non blocking mode, will return -EAGAIN.
 908  *
 909  * The btree node will have either a read or a write lock held, depending on
 910  * level and op->lock.
 911  */
 912 struct btree *bch_btree_node_get(struct cache_set *c, struct bkey *k,
 913                                  int level, struct btree_op *op)
 914 {
 915         int i = 0;
 916         bool write = level <= op->lock;
 917         struct btree *b;
 918
 919         BUG_ON(level < 0);
 920 retry:
 921         b = mca_find(c, k);
 922
 923         if (!b) {
 924                 if (current->bio_list)
 925                         return ERR_PTR(-EAGAIN);
 926
 927                 mutex_lock(&c->bucket_lock);
 928                 b = mca_alloc(c, k, level, &op->cl);
 929                 mutex_unlock(&c->bucket_lock);
 930
 931                 if (!b)
 932                         goto retry;
 933                 if (IS_ERR(b))
 934                         return b;
 935
 936                 bch_btree_node_read(b);
 937
 938                 if (!write)
 939                         downgrade_write(&b->lock);
 940         } else {
 941                 rw_lock(write, b, level);
 942                 if (PTR_HASH(c, &b->key) != PTR_HASH(c, k)) {
 943                         rw_unlock(write, b);
 944                         goto retry;
 945                 }
 946                 BUG_ON(b->level != level);
 947         }
 948
 949         b->accessed = 1;
 950
 951         for (; i <= b->nsets && b->sets[i].size; i++) {
 952                 prefetch(b->sets[i].tree);
 953                 prefetch(b->sets[i].data);
 954         }
 955
 956         for (; i <= b->nsets; i++)
 957                 prefetch(b->sets[i].data);
 958
 959         if (btree_node_io_error(b)) {
 960                 rw_unlock(write, b);
 961                 return ERR_PTR(-EIO);
 962         }
 963
 964         BUG_ON(!b->written);
 965
 966         return b;
 967 }
 968
 969 static void btree_node_prefetch(struct cache_set *c, struct bkey *k, int level)
 970 {
 971         struct btree *b;
 972
 973         mutex_lock(&c->bucket_lock);
 974         b = mca_alloc(c, k, level, NULL);
 975         mutex_unlock(&c->bucket_lock);
 976
 977         if (!IS_ERR_OR_NULL(b)) {
 978                 bch_btree_node_read(b);
 979                 rw_unlock(true, b);
 980         }
 981 }
 982
 983 /* Btree alloc */
 984
 985 static void btree_node_free(struct btree *b, struct btree_op *op)
 986 {
 987         unsigned i;
 988
 989         trace_bcache_btree_node_free(b);
 990
 991         /*
 992          * The BUG_ON() in btree_node_get() implies that we must have a write
 993          * lock on parent to free or even invalidate a node
 994          */
 995         BUG_ON(op->lock <= b->level);
 996         BUG_ON(b == b->c->root);
 997
 998         if (btree_node_dirty(b))
 999                 btree_complete_write(b, btree_current_write(b));
1000         clear_bit(BTREE_NODE_dirty, &b->flags);
1001
1002         cancel_delayed_work(&b->work);
1003
1004         mutex_lock(&b->c->bucket_lock);
1005
1006         for (i = 0; i < KEY_PTRS(&b->key); i++) {
1007                 BUG_ON(atomic_read(&PTR_BUCKET(b->c, &b->key, i)->pin));
1008
1009                 bch_inc_gen(PTR_CACHE(b->c, &b->key, i),
1010                             PTR_BUCKET(b->c, &b->key, i));
1011         }
1012
1013         bch_bucket_free(b->c, &b->key);
1014         mca_bucket_free(b);
1015         mutex_unlock(&b->c->bucket_lock);
1016 }
1017
1018 struct btree *bch_btree_node_alloc(struct cache_set *c, int level,
1019                                    struct closure *cl)
1020 {
1021         BKEY_PADDED(key) k;
1022         struct btree *b = ERR_PTR(-EAGAIN);
1023
1024         mutex_lock(&c->bucket_lock);
1025 retry:
1026         if (__bch_bucket_alloc_set(c, WATERMARK_METADATA, &k.key, 1, cl))
1027                 goto err;
1028
1029         SET_KEY_SIZE(&k.key, c->btree_pages * PAGE_SECTORS);
1030
1031         b = mca_alloc(c, &k.key, level, cl);
1032         if (IS_ERR(b))
1033                 goto err_free;
1034
1035         if (!b) {
1036                 cache_bug(c,
1037                         "Tried to allocate bucket that was in btree cache");
1038                 __bkey_put(c, &k.key);
1039                 goto retry;
1040         }
1041
1042         b->accessed = 1;
1043         bch_bset_init_next(b);
1044
1045         mutex_unlock(&c->bucket_lock);
1046
1047         trace_bcache_btree_node_alloc(b);
1048         return b;
1049 err_free:
1050         bch_bucket_free(c, &k.key);
1051         __bkey_put(c, &k.key);
1052 err:
1053         mutex_unlock(&c->bucket_lock);
1054
1055         trace_bcache_btree_node_alloc_fail(b);
1056         return b;
1057 }
1058
1059 static struct btree *btree_node_alloc_replacement(struct btree *b,
1060                                                   struct closure *cl)
1061 {
1062         struct btree *n = bch_btree_node_alloc(b->c, b->level, cl);
1063         if (!IS_ERR_OR_NULL(n))
1064                 bch_btree_sort_into(b, n);
1065
1066         return n;
1067 }
1068
1069 /* Garbage collection */
1070
1071 uint8_t __bch_btree_mark_key(struct cache_set *c, int level, struct bkey *k)
1072 {
1073         uint8_t stale = 0;
1074         unsigned i;
1075         struct bucket *g;
1076
1077         /*
1078          * ptr_invalid() can't return true for the keys that mark btree nodes as
1079          * freed, but since ptr_bad() returns true we'll never actually use them
1080          * for anything and thus we don't want mark their pointers here
1081          */
1082         if (!bkey_cmp(k, &ZERO_KEY))
1083                 return stale;
1084
1085         for (i = 0; i < KEY_PTRS(k); i++) {
1086                 if (!ptr_available(c, k, i))
1087                         continue;
1088
1089                 g = PTR_BUCKET(c, k, i);
1090
1091                 if (gen_after(g->gc_gen, PTR_GEN(k, i)))
1092                         g->gc_gen = PTR_GEN(k, i);
1093
1094                 if (ptr_stale(c, k, i)) {
1095                         stale = max(stale, ptr_stale(c, k, i));
1096                         continue;
1097                 }
1098
1099                 cache_bug_on(GC_MARK(g) &&
1100                              (GC_MARK(g) == GC_MARK_METADATA) != (level != 0),
1101                              c, "inconsistent ptrs: mark = %llu, level = %i",
1102                              GC_MARK(g), level);
1103
1104                 if (level)
1105                         SET_GC_MARK(g, GC_MARK_METADATA);
1106                 else if (KEY_DIRTY(k))
1107                         SET_GC_MARK(g, GC_MARK_DIRTY);
1108
1109                 /* guard against overflow */
1110                 SET_GC_SECTORS_USED(g, min_t(unsigned,
1111                                              GC_SECTORS_USED(g) + KEY_SIZE(k),
1112                                              (1 << 14) - 1));
1113
1114                 BUG_ON(!GC_SECTORS_USED(g));
1115         }
1116
1117         return stale;
1118 }
1119
1120 #define btree_mark_key(b, k)    __bch_btree_mark_key(b->c, b->level, k)
1121
1122 static int btree_gc_mark_node(struct btree *b, unsigned *keys,
1123                               struct gc_stat *gc)
1124 {
1125         uint8_t stale = 0;
1126         unsigned last_dev = -1;
1127         struct bcache_device *d = NULL;
1128         struct bkey *k;
1129         struct btree_iter iter;
1130         struct bset_tree *t;
1131
1132         gc->nodes++;
1133
1134         for_each_key_filter(b, k, &iter, bch_ptr_invalid) {
1135                 if (last_dev != KEY_INODE(k)) {
1136                         last_dev = KEY_INODE(k);
1137
1138                         d = KEY_INODE(k) < b->c->nr_uuids
1139                                 ? b->c->devices[last_dev]
1140                                 : NULL;
1141                 }
1142
1143                 stale = max(stale, btree_mark_key(b, k));
1144
1145                 if (bch_ptr_bad(b, k))
1146                         continue;
1147
1148                 *keys += bkey_u64s(k);
1149
1150                 gc->key_bytes += bkey_u64s(k);
1151                 gc->nkeys++;
1152
1153                 gc->data += KEY_SIZE(k);
1154                 if (KEY_DIRTY(k))
1155                         gc->dirty += KEY_SIZE(k);
1156         }
1157
1158         for (t = b->sets; t <= &b->sets[b->nsets]; t++)
1159                 btree_bug_on(t->size &&
1160                              bset_written(b, t) &&
1161                              bkey_cmp(&b->key, &t->end) < 0,
1162                              b, "found short btree key in gc");
1163
1164         return stale;
1165 }
1166
1167 static struct btree *btree_gc_alloc(struct btree *b, struct bkey *k,
1168                                     struct btree_op *op)
1169 {
1170         /*
1171          * We block priorities from being written for the duration of garbage
1172          * collection, so we can't sleep in btree_alloc() ->
1173          * bch_bucket_alloc_set(), or we'd risk deadlock - so we don't pass it
1174          * our closure.
1175          */
1176         struct btree *n = btree_node_alloc_replacement(b, NULL);
1177
1178         if (!IS_ERR_OR_NULL(n)) {
1179                 swap(b, n);
1180                 __bkey_put(b->c, &b->key);
1181
1182                 memcpy(k->ptr, b->key.ptr,
1183                        sizeof(uint64_t) * KEY_PTRS(&b->key));
1184
1185                 btree_node_free(n, op);
1186                 up_write(&n->lock);
1187         }
1188
1189         return b;
1190 }
1191
1192 /*
1193  * Leaving this at 2 until we've got incremental garbage collection done; it
1194  * could be higher (and has been tested with 4) except that garbage collection
1195  * could take much longer, adversely affecting latency.
1196  */
1197 #define GC_MERGE_NODES  2U
1198
1199 struct gc_merge_info {
1200         struct btree    *b;
1201         struct bkey     *k;
1202         unsigned        keys;
1203 };
1204
1205 static void btree_gc_coalesce(struct btree *b, struct btree_op *op,
1206                               struct gc_stat *gc, struct gc_merge_info *r)
1207 {
1208         unsigned nodes = 0, keys = 0, blocks;
1209         int i;
1210
1211         while (nodes < GC_MERGE_NODES && r[nodes].b)
1212                 keys += r[nodes++].keys;
1213
1214         blocks = btree_default_blocks(b->c) * 2 / 3;
1215
1216         if (nodes < 2 ||
1217             __set_blocks(b->sets[0].data, keys, b->c) > blocks * (nodes - 1))
1218                 return;
1219
1220         for (i = nodes - 1; i >= 0; --i) {
1221                 if (r[i].b->written)
1222                         r[i].b = btree_gc_alloc(r[i].b, r[i].k, op);
1223
1224                 if (r[i].b->written)
1225                         return;
1226         }
1227
1228         for (i = nodes - 1; i > 0; --i) {
1229                 struct bset *n1 = r[i].b->sets->data;
1230                 struct bset *n2 = r[i - 1].b->sets->data;
1231                 struct bkey *k, *last = NULL;
1232
1233                 keys = 0;
1234
1235                 if (i == 1) {
1236                         /*
1237                          * Last node we're not getting rid of - we're getting
1238                          * rid of the node at r[0]. Have to try and fit all of
1239                          * the remaining keys into this node; we can't ensure
1240                          * they will always fit due to rounding and variable
1241                          * length keys (shouldn't be possible in practice,
1242                          * though)
1243                          */
1244                         if (__set_blocks(n1, n1->keys + r->keys,
1245                                          b->c) > btree_blocks(r[i].b))
1246                                 return;
1247
1248                         keys = n2->keys;
1249                         last = &r->b->key;
1250                 } else
1251                         for (k = n2->start;
1252                              k < end(n2);
1253                              k = bkey_next(k)) {
1254                                 if (__set_blocks(n1, n1->keys + keys +
1255                                                  bkey_u64s(k), b->c) > blocks)
1256                                         break;
1257
1258                                 last = k;
1259                                 keys += bkey_u64s(k);
1260                         }
1261
1262                 BUG_ON(__set_blocks(n1, n1->keys + keys,
1263                                     b->c) > btree_blocks(r[i].b));
1264
1265                 if (last) {
1266                         bkey_copy_key(&r[i].b->key, last);
1267                         bkey_copy_key(r[i].k, last);
1268                 }
1269
1270                 memcpy(end(n1),
1271                        n2->start,
1272                        (void *) node(n2, keys) - (void *) n2->start);
1273
1274                 n1->keys += keys;
1275
1276                 memmove(n2->start,
1277                         node(n2, keys),
1278                         (void *) end(n2) - (void *) node(n2, keys));
1279
1280                 n2->keys -= keys;
1281
1282                 r[i].keys       = n1->keys;
1283                 r[i - 1].keys   = n2->keys;
1284         }
1285
1286         btree_node_free(r->b, op);
1287         up_write(&r->b->lock);
1288
1289         trace_bcache_btree_gc_coalesce(nodes);
1290
1291         gc->nodes--;
1292         nodes--;
1293
1294         memmove(&r[0], &r[1], sizeof(struct gc_merge_info) * nodes);
1295         memset(&r[nodes], 0, sizeof(struct gc_merge_info));
1296 }
1297
1298 static int btree_gc_recurse(struct btree *b, struct btree_op *op,
1299                             struct closure *writes, struct gc_stat *gc)
1300 {
1301         void write(struct btree *r)
1302         {
1303                 if (!r->written)
1304                         bch_btree_node_write(r, &op->cl);
1305                 else if (btree_node_dirty(r))
1306                         bch_btree_node_write(r, writes);
1307
1308                 up_write(&r->lock);
1309         }
1310
1311         int ret = 0, stale;
1312         unsigned i;
1313         struct gc_merge_info r[GC_MERGE_NODES];
1314
1315         memset(r, 0, sizeof(r));
1316
1317         while ((r->k = bch_next_recurse_key(b, &b->c->gc_done))) {
1318                 r->b = bch_btree_node_get(b->c, r->k, b->level - 1, op);
1319
1320                 if (IS_ERR(r->b)) {
1321                         ret = PTR_ERR(r->b);
1322                         break;
1323                 }
1324
1325                 r->keys = 0;
1326                 stale = btree_gc_mark_node(r->b, &r->keys, gc);
1327
1328                 if (!b->written &&
1329                     (r->b->level || stale > 10 ||
1330                      b->c->gc_always_rewrite))
1331                         r->b = btree_gc_alloc(r->b, r->k, op);
1332
1333                 if (r->b->level)
1334                         ret = btree_gc_recurse(r->b, op, writes, gc);
1335
1336                 if (ret) {
1337                         write(r->b);
1338                         break;
1339                 }
1340
1341                 bkey_copy_key(&b->c->gc_done, r->k);
1342
1343                 if (!b->written)
1344                         btree_gc_coalesce(b, op, gc, r);
1345
1346                 if (r[GC_MERGE_NODES - 1].b)
1347                         write(r[GC_MERGE_NODES - 1].b);
1348
1349                 memmove(&r[1], &r[0],
1350                         sizeof(struct gc_merge_info) * (GC_MERGE_NODES - 1));
1351
1352                 /* When we've got incremental GC working, we'll want to do
1353                  * if (should_resched())
1354                  *      return -EAGAIN;
1355                  */
1356                 cond_resched();
1357 #if 0
1358                 if (need_resched()) {
1359                         ret = -EAGAIN;
1360                         break;
1361                 }
1362 #endif
1363         }
1364
1365         for (i = 1; i < GC_MERGE_NODES && r[i].b; i++)
1366                 write(r[i].b);
1367
1368         /* Might have freed some children, must remove their keys */
1369         if (!b->written)
1370                 bch_btree_sort(b);
1371
1372         return ret;
1373 }
1374
1375 static int bch_btree_gc_root(struct btree *b, struct btree_op *op,
1376                              struct closure *writes, struct gc_stat *gc)
1377 {
1378         struct btree *n = NULL;
1379         unsigned keys = 0;
1380         int ret = 0, stale = btree_gc_mark_node(b, &keys, gc);
1381
1382         if (b->level || stale > 10)
1383                 n = btree_node_alloc_replacement(b, NULL);
1384
1385         if (!IS_ERR_OR_NULL(n))
1386                 swap(b, n);
1387
1388         if (b->level)
1389                 ret = btree_gc_recurse(b, op, writes, gc);
1390
1391         if (!b->written || btree_node_dirty(b)) {
1392                 bch_btree_node_write(b, n ? &op->cl : NULL);
1393         }
1394
1395         if (!IS_ERR_OR_NULL(n)) {
1396                 closure_sync(&op->cl);
1397                 bch_btree_set_root(b);
1398                 btree_node_free(n, op);
1399                 rw_unlock(true, b);
1400         }
1401
1402         return ret;
1403 }
1404
1405 static void btree_gc_start(struct cache_set *c)
1406 {
1407         struct cache *ca;
1408         struct bucket *b;
1409         unsigned i;
1410
1411         if (!c->gc_mark_valid)
1412                 return;
1413
1414         mutex_lock(&c->bucket_lock);
1415
1416         c->gc_mark_valid = 0;
1417         c->gc_done = ZERO_KEY;
1418
1419         for_each_cache(ca, c, i)
1420                 for_each_bucket(b, ca) {
1421                         b->gc_gen = b->gen;
1422                         if (!atomic_read(&b->pin)) {
1423                                 SET_GC_MARK(b, GC_MARK_RECLAIMABLE);
1424                                 SET_GC_SECTORS_USED(b, 0);
1425                         }
1426                 }
1427
1428         mutex_unlock(&c->bucket_lock);
1429 }
1430
1431 size_t bch_btree_gc_finish(struct cache_set *c)
1432 {
1433         size_t available = 0;
1434         struct bucket *b;
1435         struct cache *ca;
1436         unsigned i;
1437
1438         mutex_lock(&c->bucket_lock);
1439
1440         set_gc_sectors(c);
1441         c->gc_mark_valid = 1;
1442         c->need_gc      = 0;
1443
1444         if (c->root)
1445                 for (i = 0; i < KEY_PTRS(&c->root->key); i++)
1446                         SET_GC_MARK(PTR_BUCKET(c, &c->root->key, i),
1447                                     GC_MARK_METADATA);
1448
1449         for (i = 0; i < KEY_PTRS(&c->uuid_bucket); i++)
1450                 SET_GC_MARK(PTR_BUCKET(c, &c->uuid_bucket, i),
1451                             GC_MARK_METADATA);
1452
1453         for_each_cache(ca, c, i) {
1454                 uint64_t *i;
1455
1456                 ca->invalidate_needs_gc = 0;
1457
1458                 for (i = ca->sb.d; i < ca->sb.d + ca->sb.keys; i++)
1459                         SET_GC_MARK(ca->buckets + *i, GC_MARK_METADATA);
1460
1461                 for (i = ca->prio_buckets;
1462                      i < ca->prio_buckets + prio_buckets(ca) * 2; i++)
1463                         SET_GC_MARK(ca->buckets + *i, GC_MARK_METADATA);
1464
1465                 for_each_bucket(b, ca) {
1466                         b->last_gc      = b->gc_gen;
1467                         c->need_gc      = max(c->need_gc, bucket_gc_gen(b));
1468
1469                         if (!atomic_read(&b->pin) &&
1470                             GC_MARK(b) == GC_MARK_RECLAIMABLE) {
1471                                 available++;
1472                                 if (!GC_SECTORS_USED(b))
1473                                         bch_bucket_add_unused(ca, b);
1474                         }
1475                 }
1476         }
1477
1478         mutex_unlock(&c->bucket_lock);
1479         return available;
1480 }
1481
1482 static void bch_btree_gc(struct closure *cl)
1483 {
1484         struct cache_set *c = container_of(cl, struct cache_set, gc.cl);
1485         int ret;
1486         unsigned long available;
1487         struct gc_stat stats;
1488         struct closure writes;
1489         struct btree_op op;
1490         uint64_t start_time = local_clock();
1491
1492         trace_bcache_gc_start(c);
1493
1494         memset(&stats, 0, sizeof(struct gc_stat));
1495         closure_init_stack(&writes);
1496         bch_btree_op_init_stack(&op);
1497         op.lock = SHRT_MAX;
1498
1499         btree_gc_start(c);
1500
1501         atomic_inc(&c->prio_blocked);
1502
1503         ret = btree_root(gc_root, c, &op, &writes, &stats);
1504         closure_sync(&op.cl);
1505         closure_sync(&writes);
1506
1507         if (ret) {
1508                 pr_warn("gc failed!");
1509                 continue_at(cl, bch_btree_gc, bch_gc_wq);
1510         }
1511
1512         /* Possibly wait for new UUIDs or whatever to hit disk */
1513         bch_journal_meta(c, &op.cl);
1514         closure_sync(&op.cl);
1515
1516         available = bch_btree_gc_finish(c);
1517
1518         atomic_dec(&c->prio_blocked);
1519         wake_up_allocators(c);
1520
1521         bch_time_stats_update(&c->btree_gc_time, start_time);
1522
1523         stats.key_bytes *= sizeof(uint64_t);
1524         stats.dirty     <<= 9;
1525         stats.data      <<= 9;
1526         stats.in_use    = (c->nbuckets - available) * 100 / c->nbuckets;
1527         memcpy(&c->gc_stats, &stats, sizeof(struct gc_stat));
1528
1529         trace_bcache_gc_end(c);
1530
1531         continue_at(cl, bch_moving_gc, bch_gc_wq);
1532 }
1533
1534 void bch_queue_gc(struct cache_set *c)
1535 {
1536         closure_trylock_call(&c->gc.cl, bch_btree_gc, bch_gc_wq, &c->cl);
1537 }
1538
1539 /* Initial partial gc */
1540
1541 static int bch_btree_check_recurse(struct btree *b, struct btree_op *op,
1542                                    unsigned long **seen)
1543 {
1544         int ret;
1545         unsigned i;
1546         struct bkey *k;
1547         struct bucket *g;
1548         struct btree_iter iter;
1549
1550         for_each_key_filter(b, k, &iter, bch_ptr_invalid) {
1551                 for (i = 0; i < KEY_PTRS(k); i++) {
1552                         if (!ptr_available(b->c, k, i))
1553                                 continue;
1554
1555                         g = PTR_BUCKET(b->c, k, i);
1556
1557                         if (!__test_and_set_bit(PTR_BUCKET_NR(b->c, k, i),
1558                                                 seen[PTR_DEV(k, i)]) ||
1559                             !ptr_stale(b->c, k, i)) {
1560                                 g->gen = PTR_GEN(k, i);
1561
1562                                 if (b->level)
1563                                         g->prio = BTREE_PRIO;
1564                                 else if (g->prio == BTREE_PRIO)
1565                                         g->prio = INITIAL_PRIO;
1566                         }
1567                 }
1568
1569                 btree_mark_key(b, k);
1570         }
1571
1572         if (b->level) {
1573                 k = bch_next_recurse_key(b, &ZERO_KEY);
1574
1575                 while (k) {
1576                         struct bkey *p = bch_next_recurse_key(b, k);
1577                         if (p)
1578                                 btree_node_prefetch(b->c, p, b->level - 1);
1579
1580                         ret = btree(check_recurse, k, b, op, seen);
1581                         if (ret)
1582                                 return ret;
1583
1584                         k = p;
1585                 }
1586         }
1587
1588         return 0;
1589 }
1590
1591 int bch_btree_check(struct cache_set *c, struct btree_op *op)
1592 {
1593         int ret = -ENOMEM;
1594         unsigned i;
1595         unsigned long *seen[MAX_CACHES_PER_SET];
1596
1597         memset(seen, 0, sizeof(seen));
1598
1599         for (i = 0; c->cache[i]; i++) {
1600                 size_t n = DIV_ROUND_UP(c->cache[i]->sb.nbuckets, 8);
1601                 seen[i] = kmalloc(n, GFP_KERNEL);
1602                 if (!seen[i])
1603                         goto err;
1604
1605                 /* Disables the seen array until prio_read() uses it too */
1606                 memset(seen[i], 0xFF, n);
1607         }
1608
1609         ret = btree_root(check_recurse, c, op, seen);
1610 err:
1611         for (i = 0; i < MAX_CACHES_PER_SET; i++)
1612                 kfree(seen[i]);
1613         return ret;
1614 }
1615
1616 /* Btree insertion */
1617
1618 static void shift_keys(struct btree *b, struct bkey *where, struct bkey *insert)
1619 {
1620         struct bset *i = b->sets[b->nsets].data;
1621
1622         memmove((uint64_t *) where + bkey_u64s(insert),
1623                 where,
1624                 (void *) end(i) - (void *) where);
1625
1626         i->keys += bkey_u64s(insert);
1627         bkey_copy(where, insert);
1628         bch_bset_fix_lookup_table(b, where);
1629 }
1630
1631 static bool fix_overlapping_extents(struct btree *b,
1632                                     struct bkey *insert,
1633                                     struct btree_iter *iter,
1634                                     struct btree_op *op)
1635 {
1636         void subtract_dirty(struct bkey *k, uint64_t offset, int sectors)
1637         {
1638                 if (KEY_DIRTY(k))
1639                         bcache_dev_sectors_dirty_add(b->c, KEY_INODE(k),
1640                                                      offset, -sectors);
1641         }
1642
1643         uint64_t old_offset;
1644         unsigned old_size, sectors_found = 0;
1645
1646         while (1) {
1647                 struct bkey *k = bch_btree_iter_next(iter);
1648                 if (!k ||
1649                     bkey_cmp(&START_KEY(k), insert) >= 0)
1650                         break;
1651
1652                 if (bkey_cmp(k, &START_KEY(insert)) <= 0)
1653                         continue;
1654
1655                 old_offset = KEY_START(k);
1656                 old_size = KEY_SIZE(k);
1657
1658                 /*
1659                  * We might overlap with 0 size extents; we can't skip these
1660                  * because if they're in the set we're inserting to we have to
1661                  * adjust them so they don't overlap with the key we're
1662                  * inserting. But we don't want to check them for BTREE_REPLACE
1663                  * operations.
1664                  */
1665
1666                 if (op->type == BTREE_REPLACE &&
1667                     KEY_SIZE(k)) {
1668                         /*
1669                          * k might have been split since we inserted/found the
1670                          * key we're replacing
1671                          */
1672                         unsigned i;
1673                         uint64_t offset = KEY_START(k) -
1674                                 KEY_START(&op->replace);
1675
1676                         /* But it must be a subset of the replace key */
1677                         if (KEY_START(k) < KEY_START(&op->replace) ||
1678                             KEY_OFFSET(k) > KEY_OFFSET(&op->replace))
1679                                 goto check_failed;
1680
1681                         /* We didn't find a key that we were supposed to */
1682                         if (KEY_START(k) > KEY_START(insert) + sectors_found)
1683                                 goto check_failed;
1684
1685                         if (KEY_PTRS(&op->replace) != KEY_PTRS(k))
1686                                 goto check_failed;
1687
1688                         /* skip past gen */
1689                         offset <<= 8;
1690
1691                         BUG_ON(!KEY_PTRS(&op->replace));
1692
1693                         for (i = 0; i < KEY_PTRS(&op->replace); i++)
1694                                 if (k->ptr[i] != op->replace.ptr[i] + offset)
1695                                         goto check_failed;
1696
1697                         sectors_found = KEY_OFFSET(k) - KEY_START(insert);
1698                 }
1699
1700                 if (bkey_cmp(insert, k) < 0 &&
1701                     bkey_cmp(&START_KEY(insert), &START_KEY(k)) > 0) {
1702                         /*
1703                          * We overlapped in the middle of an existing key: that
1704                          * means we have to split the old key. But we have to do
1705                          * slightly different things depending on whether the
1706                          * old key has been written out yet.
1707                          */
1708
1709                         struct bkey *top;
1710
1711                         subtract_dirty(k, KEY_START(insert), KEY_SIZE(insert));
1712
1713                         if (bkey_written(b, k)) {
1714                                 /*
1715                                  * We insert a new key to cover the top of the
1716                                  * old key, and the old key is modified in place
1717                                  * to represent the bottom split.
1718                                  *
1719                                  * It's completely arbitrary whether the new key
1720                                  * is the top or the bottom, but it has to match
1721                                  * up with what btree_sort_fixup() does - it
1722                                  * doesn't check for this kind of overlap, it
1723                                  * depends on us inserting a new key for the top
1724                                  * here.
1725                                  */
1726                                 top = bch_bset_search(b, &b->sets[b->nsets],
1727                                                       insert);
1728                                 shift_keys(b, top, k);
1729                         } else {
1730                                 BKEY_PADDED(key) temp;
1731                                 bkey_copy(&temp.key, k);
1732                                 shift_keys(b, k, &temp.key);
1733                                 top = bkey_next(k);
1734                         }
1735
1736                         bch_cut_front(insert, top);
1737                         bch_cut_back(&START_KEY(insert), k);
1738                         bch_bset_fix_invalidated_key(b, k);
1739                         return false;
1740                 }
1741
1742                 if (bkey_cmp(insert, k) < 0) {
1743                         bch_cut_front(insert, k);
1744                 } else {
1745                         if (bkey_written(b, k) &&
1746                             bkey_cmp(&START_KEY(insert), &START_KEY(k)) <= 0) {
1747                                 /*
1748                                  * Completely overwrote, so we don't have to
1749                                  * invalidate the binary search tree
1750                                  */
1751                                 bch_cut_front(k, k);
1752                         } else {
1753                                 __bch_cut_back(&START_KEY(insert), k);
1754                                 bch_bset_fix_invalidated_key(b, k);
1755                         }
1756                 }
1757
1758                 subtract_dirty(k, old_offset, old_size - KEY_SIZE(k));
1759         }
1760
1761 check_failed:
1762         if (op->type == BTREE_REPLACE) {
1763                 if (!sectors_found) {
1764                         op->insert_collision = true;
1765                         return true;
1766                 } else if (sectors_found < KEY_SIZE(insert)) {
1767                         SET_KEY_OFFSET(insert, KEY_OFFSET(insert) -
1768                                        (KEY_SIZE(insert) - sectors_found));
1769                         SET_KEY_SIZE(insert, sectors_found);
1770                 }
1771         }
1772
1773         return false;
1774 }
1775
1776 static bool btree_insert_key(struct btree *b, struct btree_op *op,
1777                              struct bkey *k)
1778 {
1779         struct bset *i = b->sets[b->nsets].data;
1780         struct bkey *m, *prev;
1781         unsigned status = BTREE_INSERT_STATUS_INSERT;
1782
1783         BUG_ON(bkey_cmp(k, &b->key) > 0);
1784         BUG_ON(b->level && !KEY_PTRS(k));
1785         BUG_ON(!b->level && !KEY_OFFSET(k));
1786
1787         if (!b->level) {
1788                 struct btree_iter iter;
1789                 struct bkey search = KEY(KEY_INODE(k), KEY_START(k), 0);
1790
1791                 /*
1792                  * bset_search() returns the first key that is strictly greater
1793                  * than the search key - but for back merging, we want to find
1794                  * the first key that is greater than or equal to KEY_START(k) -
1795                  * unless KEY_START(k) is 0.
1796                  */
1797                 if (KEY_OFFSET(&search))
1798                         SET_KEY_OFFSET(&search, KEY_OFFSET(&search) - 1);
1799
1800                 prev = NULL;
1801                 m = bch_btree_iter_init(b, &iter, &search);
1802
1803                 if (fix_overlapping_extents(b, k, &iter, op))
1804                         return false;
1805
1806                 while (m != end(i) &&
1807                        bkey_cmp(k, &START_KEY(m)) > 0)
1808                         prev = m, m = bkey_next(m);
1809
1810                 if (key_merging_disabled(b->c))
1811                         goto insert;
1812
1813                 /* prev is in the tree, if we merge we're done */
1814                 status = BTREE_INSERT_STATUS_BACK_MERGE;
1815                 if (prev &&
1816                     bch_bkey_try_merge(b, prev, k))
1817                         goto merged;
1818
1819                 status = BTREE_INSERT_STATUS_OVERWROTE;
1820                 if (m != end(i) &&
1821                     KEY_PTRS(m) == KEY_PTRS(k) && !KEY_SIZE(m))
1822                         goto copy;
1823
1824                 status = BTREE_INSERT_STATUS_FRONT_MERGE;
1825                 if (m != end(i) &&
1826                     bch_bkey_try_merge(b, k, m))
1827                         goto copy;
1828         } else
1829                 m = bch_bset_search(b, &b->sets[b->nsets], k);
1830
1831 insert: shift_keys(b, m, k);
1832 copy:   bkey_copy(m, k);
1833 merged:
1834         if (KEY_DIRTY(k))
1835                 bcache_dev_sectors_dirty_add(b->c, KEY_INODE(k),
1836                                              KEY_START(k), KEY_SIZE(k));
1837
1838         bch_check_keys(b, "%u for %s", status, op_type(op));
1839
1840         if (b->level && !KEY_OFFSET(k))
1841                 btree_current_write(b)->prio_blocked++;
1842
1843         trace_bcache_btree_insert_key(b, k, op->type, status);
1844
1845         return true;
1846 }
1847
1848 static bool bch_btree_insert_keys(struct btree *b, struct btree_op *op)
1849 {
1850         bool ret = false;
1851         struct bkey *k;
1852         unsigned oldsize = bch_count_data(b);
1853
1854         while ((k = bch_keylist_pop(&op->keys))) {
1855                 bkey_put(b->c, k, b->level);
1856                 ret |= btree_insert_key(b, op, k);
1857         }
1858
1859         BUG_ON(bch_count_data(b) < oldsize);
1860         return ret;
1861 }
1862
1863 bool bch_btree_insert_check_key(struct btree *b, struct btree_op *op,
1864                                    struct bio *bio)
1865 {
1866         bool ret = false;
1867         uint64_t btree_ptr = b->key.ptr[0];
1868         unsigned long seq = b->seq;
1869         BKEY_PADDED(k) tmp;
1870
1871         rw_unlock(false, b);
1872         rw_lock(true, b, b->level);
1873
1874         if (b->key.ptr[0] != btree_ptr ||
1875             b->seq != seq + 1 ||
1876             should_split(b))
1877                 goto out;
1878
1879         op->replace = KEY(op->inode, bio_end_sector(bio), bio_sectors(bio));
1880
1881         SET_KEY_PTRS(&op->replace, 1);
1882         get_random_bytes(&op->replace.ptr[0], sizeof(uint64_t));
1883
1884         SET_PTR_DEV(&op->replace, 0, PTR_CHECK_DEV);
1885
1886         bkey_copy(&tmp.k, &op->replace);
1887
1888         BUG_ON(op->type != BTREE_INSERT);
1889         BUG_ON(!btree_insert_key(b, op, &tmp.k));
1890         ret = true;
1891 out:
1892         downgrade_write(&b->lock);
1893         return ret;
1894 }
1895
1896 static int btree_split(struct btree *b, struct btree_op *op)
1897 {
1898         bool split, root = b == b->c->root;
1899         struct btree *n1, *n2 = NULL, *n3 = NULL;
1900         uint64_t start_time = local_clock();
1901
1902         if (b->level)
1903                 set_closure_blocking(&op->cl);
1904
1905         n1 = btree_node_alloc_replacement(b, &op->cl);
1906         if (IS_ERR(n1))
1907                 goto err;
1908
1909         split = set_blocks(n1->sets[0].data, n1->c) > (btree_blocks(b) * 4) / 5;
1910
1911         if (split) {
1912                 unsigned keys = 0;
1913
1914                 trace_bcache_btree_node_split(b, n1->sets[0].data->keys);
1915
1916                 n2 = bch_btree_node_alloc(b->c, b->level, &op->cl);
1917                 if (IS_ERR(n2))
1918                         goto err_free1;
1919
1920                 if (root) {
1921                         n3 = bch_btree_node_alloc(b->c, b->level + 1, &op->cl);
1922                         if (IS_ERR(n3))
1923                                 goto err_free2;
1924                 }
1925
1926                 bch_btree_insert_keys(n1, op);
1927
1928                 /* Has to be a linear search because we don't have an auxiliary
1929                  * search tree yet
1930                  */
1931
1932                 while (keys < (n1->sets[0].data->keys * 3) / 5)
1933                         keys += bkey_u64s(node(n1->sets[0].data, keys));
1934
1935                 bkey_copy_key(&n1->key, node(n1->sets[0].data, keys));
1936                 keys += bkey_u64s(node(n1->sets[0].data, keys));
1937
1938                 n2->sets[0].data->keys = n1->sets[0].data->keys - keys;
1939                 n1->sets[0].data->keys = keys;
1940
1941                 memcpy(n2->sets[0].data->start,
1942                        end(n1->sets[0].data),
1943                        n2->sets[0].data->keys * sizeof(uint64_t));
1944
1945                 bkey_copy_key(&n2->key, &b->key);
1946
1947                 bch_keylist_add(&op->keys, &n2->key);
1948                 bch_btree_node_write(n2, &op->cl);
1949                 rw_unlock(true, n2);
1950         } else {
1951                 trace_bcache_btree_node_compact(b, n1->sets[0].data->keys);
1952
1953                 bch_btree_insert_keys(n1, op);
1954         }
1955
1956         bch_keylist_add(&op->keys, &n1->key);
1957         bch_btree_node_write(n1, &op->cl);
1958
1959         if (n3) {
1960                 bkey_copy_key(&n3->key, &MAX_KEY);
1961                 bch_btree_insert_keys(n3, op);
1962                 bch_btree_node_write(n3, &op->cl);
1963
1964                 closure_sync(&op->cl);
1965                 bch_btree_set_root(n3);
1966                 rw_unlock(true, n3);
1967         } else if (root) {
1968                 op->keys.top = op->keys.bottom;
1969                 closure_sync(&op->cl);
1970                 bch_btree_set_root(n1);
1971         } else {
1972                 unsigned i;
1973
1974                 bkey_copy(op->keys.top, &b->key);
1975                 bkey_copy_key(op->keys.top, &ZERO_KEY);
1976
1977                 for (i = 0; i < KEY_PTRS(&b->key); i++) {
1978                         uint8_t g = PTR_BUCKET(b->c, &b->key, i)->gen + 1;
1979
1980                         SET_PTR_GEN(op->keys.top, i, g);
1981                 }
1982
1983                 bch_keylist_push(&op->keys);
1984                 closure_sync(&op->cl);
1985                 atomic_inc(&b->c->prio_blocked);
1986         }
1987
1988         rw_unlock(true, n1);
1989         btree_node_free(b, op);
1990
1991         bch_time_stats_update(&b->c->btree_split_time, start_time);
1992
1993         return 0;
1994 err_free2:
1995         __bkey_put(n2->c, &n2->key);
1996         btree_node_free(n2, op);
1997         rw_unlock(true, n2);
1998 err_free1:
1999         __bkey_put(n1->c, &n1->key);
2000         btree_node_free(n1, op);
2001         rw_unlock(true, n1);
2002 err:
2003         if (n3 == ERR_PTR(-EAGAIN) ||
2004             n2 == ERR_PTR(-EAGAIN) ||
2005             n1 == ERR_PTR(-EAGAIN))
2006                 return -EAGAIN;
2007
2008         pr_warn("couldn't split");
2009         return -ENOMEM;
2010 }
2011
2012 static int bch_btree_insert_recurse(struct btree *b, struct btree_op *op,
2013                                     struct keylist *stack_keys)
2014 {
2015         if (b->level) {
2016                 int ret;
2017                 struct bkey *insert = op->keys.bottom;
2018                 struct bkey *k = bch_next_recurse_key(b, &START_KEY(insert));
2019
2020                 if (!k) {
2021                         btree_bug(b, "no key to recurse on at level %i/%i",
2022                                   b->level, b->c->root->level);
2023
2024                         op->keys.top = op->keys.bottom;
2025                         return -EIO;
2026                 }
2027
2028                 if (bkey_cmp(insert, k) > 0) {
2029                         unsigned i;
2030
2031                         if (op->type == BTREE_REPLACE) {
2032                                 __bkey_put(b->c, insert);
2033                                 op->keys.top = op->keys.bottom;
2034                                 op->insert_collision = true;
2035                                 return 0;
2036                         }
2037
2038                         for (i = 0; i < KEY_PTRS(insert); i++)
2039                                 atomic_inc(&PTR_BUCKET(b->c, insert, i)->pin);
2040
2041                         bkey_copy(stack_keys->top, insert);
2042
2043                         bch_cut_back(k, insert);
2044                         bch_cut_front(k, stack_keys->top);
2045
2046                         bch_keylist_push(stack_keys);
2047                 }
2048
2049                 ret = btree(insert_recurse, k, b, op, stack_keys);
2050                 if (ret)
2051                         return ret;
2052         }
2053
2054         if (!bch_keylist_empty(&op->keys)) {
2055                 if (should_split(b)) {
2056                         if (op->lock <= b->c->root->level) {
2057                                 BUG_ON(b->level);
2058                                 op->lock = b->c->root->level + 1;
2059                                 return -EINTR;
2060                         }
2061                         return btree_split(b, op);
2062                 }
2063
2064                 BUG_ON(write_block(b) != b->sets[b->nsets].data);
2065
2066                 if (bch_btree_insert_keys(b, op)) {
2067                         if (!b->level)
2068                                 bch_btree_leaf_dirty(b, op);
2069                         else
2070                                 bch_btree_node_write(b, &op->cl);
2071                 }
2072         }
2073
2074         return 0;
2075 }
2076
2077 int bch_btree_insert(struct btree_op *op, struct cache_set *c)
2078 {
2079         int ret = 0;
2080         struct keylist stack_keys;
2081
2082         /*
2083          * Don't want to block with the btree locked unless we have to,
2084          * otherwise we get deadlocks with try_harder and between split/gc
2085          */
2086         clear_closure_blocking(&op->cl);
2087
2088         BUG_ON(bch_keylist_empty(&op->keys));
2089         bch_keylist_copy(&stack_keys, &op->keys);
2090         bch_keylist_init(&op->keys);
2091
2092         while (!bch_keylist_empty(&stack_keys) ||
2093                !bch_keylist_empty(&op->keys)) {
2094                 if (bch_keylist_empty(&op->keys)) {
2095                         bch_keylist_add(&op->keys,
2096                                         bch_keylist_pop(&stack_keys));
2097                         op->lock = 0;
2098                 }
2099
2100                 ret = btree_root(insert_recurse, c, op, &stack_keys);
2101
2102                 if (ret == -EAGAIN) {
2103                         ret = 0;
2104                         closure_sync(&op->cl);
2105                 } else if (ret) {
2106                         struct bkey *k;
2107
2108                         pr_err("error %i trying to insert key for %s",
2109                                ret, op_type(op));
2110
2111                         while ((k = bch_keylist_pop(&stack_keys) ?:
2112                                     bch_keylist_pop(&op->keys)))
2113                                 bkey_put(c, k, 0);
2114                 }
2115         }
2116
2117         bch_keylist_free(&stack_keys);
2118
2119         if (op->journal)
2120                 atomic_dec_bug(op->journal);
2121         op->journal = NULL;
2122         return ret;
2123 }
2124
2125 void bch_btree_set_root(struct btree *b)
2126 {
2127         unsigned i;
2128         struct closure cl;
2129
2130         closure_init_stack(&cl);
2131
2132         trace_bcache_btree_set_root(b);
2133
2134         BUG_ON(!b->written);
2135
2136         for (i = 0; i < KEY_PTRS(&b->key); i++)
2137                 BUG_ON(PTR_BUCKET(b->c, &b->key, i)->prio != BTREE_PRIO);
2138
2139         mutex_lock(&b->c->bucket_lock);
2140         list_del_init(&b->list);
2141         mutex_unlock(&b->c->bucket_lock);
2142
2143         b->c->root = b;
2144         __bkey_put(b->c, &b->key);
2145
2146         bch_journal_meta(b->c, &cl);
2147         closure_sync(&cl);
2148 }
2149
2150 /* Cache lookup */
2151
2152 static int submit_partial_cache_miss(struct btree *b, struct btree_op *op,
2153                                      struct bkey *k)
2154 {
2155         struct search *s = container_of(op, struct search, op);
2156         struct bio *bio = &s->bio.bio;
2157         int ret = 0;
2158
2159         while (!ret &&
2160                !op->lookup_done) {
2161                 unsigned sectors = INT_MAX;
2162
2163                 if (KEY_INODE(k) == op->inode) {
2164                         if (KEY_START(k) <= bio->bi_sector)
2165                                 break;
2166
2167                         sectors = min_t(uint64_t, sectors,
2168                                         KEY_START(k) - bio->bi_sector);
2169                 }
2170
2171                 ret = s->d->cache_miss(b, s, bio, sectors);
2172         }
2173
2174         return ret;
2175 }
2176
2177 /*
2178  * Read from a single key, handling the initial cache miss if the key starts in
2179  * the middle of the bio
2180  */
2181 static int submit_partial_cache_hit(struct btree *b, struct btree_op *op,
2182                                     struct bkey *k)
2183 {
2184         struct search *s = container_of(op, struct search, op);
2185         struct bio *bio = &s->bio.bio;
2186         unsigned ptr;
2187         struct bio *n;
2188
2189         int ret = submit_partial_cache_miss(b, op, k);
2190         if (ret || op->lookup_done)
2191                 return ret;
2192
2193         /* XXX: figure out best pointer - for multiple cache devices */
2194         ptr = 0;
2195
2196         PTR_BUCKET(b->c, k, ptr)->prio = INITIAL_PRIO;
2197
2198         while (!op->lookup_done &&
2199                KEY_INODE(k) == op->inode &&
2200                bio->bi_sector < KEY_OFFSET(k)) {
2201                 struct bkey *bio_key;
2202                 sector_t sector = PTR_OFFSET(k, ptr) +
2203                         (bio->bi_sector - KEY_START(k));
2204                 unsigned sectors = min_t(uint64_t, INT_MAX,
2205                                          KEY_OFFSET(k) - bio->bi_sector);
2206
2207                 n = bch_bio_split(bio, sectors, GFP_NOIO, s->d->bio_split);
2208                 if (n == bio)
2209                         op->lookup_done = true;
2210
2211                 bio_key = &container_of(n, struct bbio, bio)->key;
2212
2213                 /*
2214                  * The bucket we're reading from might be reused while our bio
2215                  * is in flight, and we could then end up reading the wrong
2216                  * data.
2217                  *
2218                  * We guard against this by checking (in cache_read_endio()) if
2219                  * the pointer is stale again; if so, we treat it as an error
2220                  * and reread from the backing device (but we don't pass that
2221                  * error up anywhere).
2222                  */
2223
2224                 bch_bkey_copy_single_ptr(bio_key, k, ptr);
2225                 SET_PTR_OFFSET(bio_key, 0, sector);
2226
2227                 n->bi_end_io    = bch_cache_read_endio;
2228                 n->bi_private   = &s->cl;
2229
2230                 __bch_submit_bbio(n, b->c);
2231         }
2232
2233         return 0;
2234 }
2235
2236 int bch_btree_search_recurse(struct btree *b, struct btree_op *op)
2237 {
2238         struct search *s = container_of(op, struct search, op);
2239         struct bio *bio = &s->bio.bio;
2240
2241         int ret = 0;
2242         struct bkey *k;
2243         struct btree_iter iter;
2244         bch_btree_iter_init(b, &iter, &KEY(op->inode, bio->bi_sector, 0));
2245
2246         do {
2247                 k = bch_btree_iter_next_filter(&iter, b, bch_ptr_bad);
2248                 if (!k) {
2249                         /*
2250                          * b->key would be exactly what we want, except that
2251                          * pointers to btree nodes have nonzero size - we
2252                          * wouldn't go far enough
2253                          */
2254
2255                         ret = submit_partial_cache_miss(b, op,
2256                                         &KEY(KEY_INODE(&b->key),
2257                                              KEY_OFFSET(&b->key), 0));
2258                         break;
2259                 }
2260
2261                 ret = b->level
2262                         ? btree(search_recurse, k, b, op)
2263                         : submit_partial_cache_hit(b, op, k);
2264         } while (!ret &&
2265                  !op->lookup_done);
2266
2267         return ret;
2268 }
2269
2270 /* Keybuf code */
2271
2272 static inline int keybuf_cmp(struct keybuf_key *l, struct keybuf_key *r)
2273 {
2274         /* Overlapping keys compare equal */
2275         if (bkey_cmp(&l->key, &START_KEY(&r->key)) <= 0)
2276                 return -1;
2277         if (bkey_cmp(&START_KEY(&l->key), &r->key) >= 0)
2278                 return 1;
2279         return 0;
2280 }
2281
2282 static inline int keybuf_nonoverlapping_cmp(struct keybuf_key *l,
2283                                             struct keybuf_key *r)
2284 {
2285         return clamp_t(int64_t, bkey_cmp(&l->key, &r->key), -1, 1);
2286 }
2287
2288 static int bch_btree_refill_keybuf(struct btree *b, struct btree_op *op,
2289                                    struct keybuf *buf, struct bkey *end,
2290                                    keybuf_pred_fn *pred)
2291 {
2292         struct btree_iter iter;
2293         bch_btree_iter_init(b, &iter, &buf->last_scanned);
2294
2295         while (!array_freelist_empty(&buf->freelist)) {
2296                 struct bkey *k = bch_btree_iter_next_filter(&iter, b,
2297                                                             bch_ptr_bad);
2298
2299                 if (!b->level) {
2300                         if (!k) {
2301                                 buf->last_scanned = b->key;
2302                                 break;
2303                         }
2304
2305                         buf->last_scanned = *k;
2306                         if (bkey_cmp(&buf->last_scanned, end) >= 0)
2307                                 break;
2308
2309                         if (pred(buf, k)) {
2310                                 struct keybuf_key *w;
2311
2312                                 spin_lock(&buf->lock);
2313
2314                                 w = array_alloc(&buf->freelist);
2315
2316                                 w->private = NULL;
2317                                 bkey_copy(&w->key, k);
2318
2319                                 if (RB_INSERT(&buf->keys, w, node, keybuf_cmp))
2320                                         array_free(&buf->freelist, w);
2321
2322                                 spin_unlock(&buf->lock);
2323                         }
2324                 } else {
2325                         if (!k)
2326                                 break;
2327
2328                         btree(refill_keybuf, k, b, op, buf, end, pred);
2329                         /*
2330                          * Might get an error here, but can't really do anything
2331                          * and it'll get logged elsewhere. Just read what we
2332                          * can.
2333                          */
2334
2335                         if (bkey_cmp(&buf->last_scanned, end) >= 0)
2336                                 break;
2337
2338                         cond_resched();
2339                 }
2340         }
2341
2342         return 0;
2343 }
2344
2345 void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf,
2346                        struct bkey *end, keybuf_pred_fn *pred)
2347 {
2348         struct bkey start = buf->last_scanned;
2349         struct btree_op op;
2350         bch_btree_op_init_stack(&op);
2351
2352         cond_resched();
2353
2354         btree_root(refill_keybuf, c, &op, buf, end, pred);
2355         closure_sync(&op.cl);
2356
2357         pr_debug("found %s keys from %llu:%llu to %llu:%llu",
2358                  RB_EMPTY_ROOT(&buf->keys) ? "no" :
2359                  array_freelist_empty(&buf->freelist) ? "some" : "a few",
2360                  KEY_INODE(&start), KEY_OFFSET(&start),
2361                  KEY_INODE(&buf->last_scanned), KEY_OFFSET(&buf->last_scanned));
2362
2363         spin_lock(&buf->lock);
2364
2365         if (!RB_EMPTY_ROOT(&buf->keys)) {
2366                 struct keybuf_key *w;
2367                 w = RB_FIRST(&buf->keys, struct keybuf_key, node);
2368                 buf->start      = START_KEY(&w->key);
2369
2370                 w = RB_LAST(&buf->keys, struct keybuf_key, node);
2371                 buf->end        = w->key;
2372         } else {
2373                 buf->start      = MAX_KEY;
2374                 buf->end        = MAX_KEY;
2375         }
2376
2377         spin_unlock(&buf->lock);
2378 }
2379
2380 static void __bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w)
2381 {
2382         rb_erase(&w->node, &buf->keys);
2383         array_free(&buf->freelist, w);
2384 }
2385
2386 void bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w)
2387 {
2388         spin_lock(&buf->lock);
2389         __bch_keybuf_del(buf, w);
2390         spin_unlock(&buf->lock);
2391 }
2392
2393 bool bch_keybuf_check_overlapping(struct keybuf *buf, struct bkey *start,
2394                                   struct bkey *end)
2395 {
2396         bool ret = false;
2397         struct keybuf_key *p, *w, s;
2398         s.key = *start;
2399
2400         if (bkey_cmp(end, &buf->start) <= 0 ||
2401             bkey_cmp(start, &buf->end) >= 0)
2402                 return false;
2403
2404         spin_lock(&buf->lock);
2405         w = RB_GREATER(&buf->keys, s, node, keybuf_nonoverlapping_cmp);
2406
2407         while (w && bkey_cmp(&START_KEY(&w->key), end) < 0) {
2408                 p = w;
2409                 w = RB_NEXT(w, node);
2410
2411                 if (p->private)
2412                         ret = true;
2413                 else
2414                         __bch_keybuf_del(buf, p);
2415         }
2416
2417         spin_unlock(&buf->lock);
2418         return ret;
2419 }
2420
2421 struct keybuf_key *bch_keybuf_next(struct keybuf *buf)
2422 {
2423         struct keybuf_key *w;
2424         spin_lock(&buf->lock);
2425
2426         w = RB_FIRST(&buf->keys, struct keybuf_key, node);
2427
2428         while (w && w->private)
2429                 w = RB_NEXT(w, node);
2430
2431         if (w)
2432                 w->private = ERR_PTR(-EINTR);
2433
2434         spin_unlock(&buf->lock);
2435         return w;
2436 }
2437
2438 struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c,
2439                                              struct keybuf *buf,
2440                                              struct bkey *end,
2441                                              keybuf_pred_fn *pred)
2442 {
2443         struct keybuf_key *ret;
2444
2445         while (1) {
2446                 ret = bch_keybuf_next(buf);
2447                 if (ret)
2448                         break;
2449
2450                 if (bkey_cmp(&buf->last_scanned, end) >= 0) {
2451                         pr_debug("scan finished");
2452                         break;
2453                 }
2454
2455                 bch_refill_keybuf(c, buf, end, pred);
2456         }
2457
2458         return ret;
2459 }
2460
2461 void bch_keybuf_init(struct keybuf *buf)
2462 {
2463         buf->last_scanned       = MAX_KEY;
2464         buf->keys               = RB_ROOT;
2465
2466         spin_lock_init(&buf->lock);
2467         array_allocator_init(&buf->freelist);
2468 }
2469
2470 void bch_btree_exit(void)
2471 {
2472         if (btree_io_wq)
2473                 destroy_workqueue(btree_io_wq);
2474         if (bch_gc_wq)
2475                 destroy_workqueue(bch_gc_wq);
2476 }
2477
2478 int __init bch_btree_init(void)
2479 {
2480         if (!(bch_gc_wq = create_singlethread_workqueue("bch_btree_gc")) ||
2481             !(btree_io_wq = create_singlethread_workqueue("bch_btree_io")))
2482                 return -ENOMEM;
2483
2484         return 0;
2485 }